Autonomous Ground Vehicle Path Following by Combining Feedback Linearization with Model Predictive Control
This thesis presents the formulation and experimental validation of an autonomous ground vehicle path following control method that combines feedback linearization (FBL) with model predictive control (MPC). The use of MPC for ground vehicle path following tasks has shown to provide high-performance results by leveraging a predictive model to better inform real-time control actions. Traditionally, MPC for ground vehicle path following requires solving computationally expensive nonlinear optimization problems. This increased computational burden can be a significant challenge for real-time implementation. This thesis presents a formulation combining MPC and FBL (MPC+FBL) that is designed to not require the use of expensive nonlinear optimization techniques to solve the MPC optimization. The result is a simple and computationally inexpensive control method that can be used on systems with limited computational resources to achieve near-optimal path following. This MPC+FBL controller was initially tested in simulation, followed by hardware experiments using a 60 kg skid-steer robot. Hardware experiments were performed indoors and outdoors on various terrains such as plastic floor tiles, grass and sand. The results from these experiments show that the MPC+FBL formulation is capable of near-optimal path following control and achieved more than 30% lower average path following errors in comparison to an implementation of a nonlinear MPC presented in the recent literature. Furthermore, this MPC+FBL formulation can complete controller calculations approximately five times faster than the implemented nonlinear MPC, validating its ability to perform model predictive control with a significantly reduced computational expense.